fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/fdtable.h>
  28 #include <linux/mm.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/swap.h>
  32 #include <linux/string.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/perf_event.h>
  36 #include <linux/highmem.h>
  37 #include <linux/spinlock.h>
  38 #include <linux/key.h>
  39 #include <linux/personality.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/utsname.h>
  42 #include <linux/pid_namespace.h>
  43 #include <linux/module.h>
  44 #include <linux/namei.h>
  45 #include <linux/proc_fs.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/tsacct_kern.h>
  50 #include <linux/cn_proc.h>
  51 #include <linux/audit.h>
  52 #include <linux/tracehook.h>
  53 #include <linux/kmod.h>
  54 #include <linux/fsnotify.h>
  55 #include <linux/fs_struct.h>
  56 #include <linux/pipe_fs_i.h>
  57
  58 #include <asm/uaccess.h>
  59 #include <asm/mmu_context.h>
  60 #include <asm/tlb.h>
  61 #include "internal.h"
  62
  63 int core_uses_pid;
  64 char core_pattern[CORENAME_MAX_SIZE] = "core";
  65 unsigned int core_pipe_limit;
  66 int suid_dumpable = 0;
  67
  68 /* The maximal length of core_pattern is also specified in sysctl.c */
  69
  70 static LIST_HEAD(formats);
  71 static DEFINE_RWLOCK(binfmt_lock);
  72
  73 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  74 {
  75         if (!fmt)
  76                 return -EINVAL;
  77         write_lock(&binfmt_lock);
  78         insert ? list_add(&fmt->lh, &formats) :
  79                  list_add_tail(&fmt->lh, &formats);
  80         write_unlock(&binfmt_lock);
  81         return 0;
  82 }
  83
  84 EXPORT_SYMBOL(__register_binfmt);
  85
  86 void unregister_binfmt(struct linux_binfmt * fmt)
  87 {
  88         write_lock(&binfmt_lock);
  89         list_del(&fmt->lh);
  90         write_unlock(&binfmt_lock);
  91 }
  92
  93 EXPORT_SYMBOL(unregister_binfmt);
  94
  95 static inline void put_binfmt(struct linux_binfmt * fmt)
  96 {
  97         module_put(fmt->module);
  98 }
  99
 100 /*
 101  * Note that a shared library must be both readable and executable due to
 102  * security reasons.
 103  *
 104  * Also note that we take the address to load from from the file itself.
 105  */
 106 SYSCALL_DEFINE1(uselib, const char __user *, library)
 107 {
 108         struct file *file;
 109         char *tmp = getname(library);
 110         int error = PTR_ERR(tmp);
 111
 112         if (IS_ERR(tmp))
 113                 goto out;
 114
 115         file = do_filp_open(AT_FDCWD, tmp,
 116                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 117                                 MAY_READ | MAY_EXEC | MAY_OPEN);
 118         putname(tmp);
 119         error = PTR_ERR(file);
 120         if (IS_ERR(file))
 121                 goto out;
 122
 123         error = -EINVAL;
 124         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 125                 goto exit;
 126
 127         error = -EACCES;
 128         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 129                 goto exit;
 130
 131         fsnotify_open(file);
 132
 133         error = -ENOEXEC;
 134         if(file->f_op) {
 135                 struct linux_binfmt * fmt;
 136
 137                 read_lock(&binfmt_lock);
 138                 list_for_each_entry(fmt, &formats, lh) {
 139                         if (!fmt->load_shlib)
 140                                 continue;
 141                         if (!try_module_get(fmt->module))
 142                                 continue;
 143                         read_unlock(&binfmt_lock);
 144                         error = fmt->load_shlib(file);
 145                         read_lock(&binfmt_lock);
 146                         put_binfmt(fmt);
 147                         if (error != -ENOEXEC)
 148                                 break;
 149                 }
 150                 read_unlock(&binfmt_lock);
 151         }
 152 exit:
 153         fput(file);
 154 out:
 155         return error;
 156 }
 157
 158 #ifdef CONFIG_MMU
 159
 160 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 161                 int write)
 162 {
 163         struct page *page;
 164         int ret;
 165
 166 #ifdef CONFIG_STACK_GROWSUP
 167         if (write) {
 168                 ret = expand_stack_downwards(bprm->vma, pos);
 169                 if (ret < 0)
 170                         return NULL;
 171         }
 172 #endif
 173         ret = get_user_pages(current, bprm->mm, pos,
 174                         1, write, 1, &page, NULL);
 175         if (ret <= 0)
 176                 return NULL;
 177
 178         if (write) {
 179                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 180                 struct rlimit *rlim;
 181
 182                 /*
 183                  * We've historically supported up to 32 pages (ARG_MAX)
 184                  * of argument strings even with small stacks
 185                  */
 186                 if (size <= ARG_MAX)
 187                         return page;
 188
 189                 /*
 190                  * Limit to 1/4-th the stack size for the argv+env strings.
 191                  * This ensures that:
 192                  *  - the remaining binfmt code will not run out of stack space,
 193                  *  - the program will have a reasonable amount of stack left
 194                  *    to work from.
 195                  */
 196                 rlim = current->signal->rlim;
 197                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 198                         put_page(page);
 199                         return NULL;
 200                 }
 201         }
 202
 203         return page;
 204 }
 205
 206 static void put_arg_page(struct page *page)
 207 {
 208         put_page(page);
 209 }
 210
 211 static void free_arg_page(struct linux_binprm *bprm, int i)
 212 {
 213 }
 214
 215 static void free_arg_pages(struct linux_binprm *bprm)
 216 {
 217 }
 218
 219 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 220                 struct page *page)
 221 {
 222         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 223 }
 224
 225 static int __bprm_mm_init(struct linux_binprm *bprm)
 226 {
 227         int err;
 228         struct vm_area_struct *vma = NULL;
 229         struct mm_struct *mm = bprm->mm;
 230
 231         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 232         if (!vma)
 233                 return -ENOMEM;
 234
 235         down_write(&mm->mmap_sem);
 236         vma->vm_mm = mm;
 237
 238         /*
 239          * Place the stack at the largest stack address the architecture
 240          * supports. Later, we'll move this to an appropriate place. We don't
 241          * use STACK_TOP because that can depend on attributes which aren't
 242          * configured yet.
 243          */
 244         BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 245         vma->vm_end = STACK_TOP_MAX;
 246         vma->vm_start = vma->vm_end - PAGE_SIZE;
 247         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 248         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 249         INIT_LIST_HEAD(&vma->anon_vma_chain);
 250         err = insert_vm_struct(mm, vma);
 251         if (err)
 252                 goto err;
 253
 254         mm->stack_vm = mm->total_vm = 1;
 255         up_write(&mm->mmap_sem);
 256         bprm->p = vma->vm_end - sizeof(void *);
 257         return 0;
 258 err:
 259         up_write(&mm->mmap_sem);
 260         bprm->vma = NULL;
 261         kmem_cache_free(vm_area_cachep, vma);
 262         return err;
 263 }
 264
 265 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 266 {
 267         return len <= MAX_ARG_STRLEN;
 268 }
 269
 270 #else
 271
 272 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 273                 int write)
 274 {
 275         struct page *page;
 276
 277         page = bprm->page[pos / PAGE_SIZE];
 278         if (!page && write) {
 279                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 280                 if (!page)
 281                         return NULL;
 282                 bprm->page[pos / PAGE_SIZE] = page;
 283         }
 284
 285         return page;
 286 }
 287
 288 static void put_arg_page(struct page *page)
 289 {
 290 }
 291
 292 static void free_arg_page(struct linux_binprm *bprm, int i)
 293 {
 294         if (bprm->page[i]) {
 295                 __free_page(bprm->page[i]);
 296                 bprm->page[i] = NULL;
 297         }
 298 }
 299
 300 static void free_arg_pages(struct linux_binprm *bprm)
 301 {
 302         int i;
 303
 304         for (i = 0; i < MAX_ARG_PAGES; i++)
 305                 free_arg_page(bprm, i);
 306 }
 307
 308 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 309                 struct page *page)
 310 {
 311 }
 312
 313 static int __bprm_mm_init(struct linux_binprm *bprm)
 314 {
 315         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 316         return 0;
 317 }
 318
 319 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 320 {
 321         return len <= bprm->p;
 322 }
 323
 324 #endif /* CONFIG_MMU */
 325
 326 /*
 327  * Create a new mm_struct and populate it with a temporary stack
 328  * vm_area_struct.  We don't have enough context at this point to set the stack
 329  * flags, permissions, and offset, so we use temporary values.  We'll update
 330  * them later in setup_arg_pages().
 331  */
 332 int bprm_mm_init(struct linux_binprm *bprm)
 333 {
 334         int err;
 335         struct mm_struct *mm = NULL;
 336
 337         bprm->mm = mm = mm_alloc();
 338         err = -ENOMEM;
 339         if (!mm)
 340                 goto err;
 341
 342         err = init_new_context(current, mm);
 343         if (err)
 344                 goto err;
 345
 346         err = __bprm_mm_init(bprm);
 347         if (err)
 348                 goto err;
 349
 350         return 0;
 351
 352 err:
 353         if (mm) {
 354                 bprm->mm = NULL;
 355                 mmdrop(mm);
 356         }
 357
 358         return err;
 359 }
 360
 361 /*
 362  * count() counts the number of strings in array ARGV.
 363  */
 364 static int count(const char __user * const __user * argv, int max)
 365 {
 366         int i = 0;
 367
 368         if (argv != NULL) {
 369                 for (;;) {
 370                         const char __user * p;
 371
 372                         if (get_user(p, argv))
 373                                 return -EFAULT;
 374                         if (!p)
 375                                 break;
 376                         argv++;
 377                         if (i++ >= max)
 378                                 return -E2BIG;
 379                         cond_resched();
 380                 }
 381         }
 382         return i;
 383 }
 384
 385 /*
 386  * 'copy_strings()' copies argument/environment strings from the old
 387  * processes's memory to the new process's stack.  The call to get_user_pages()
 388  * ensures the destination page is created and not swapped out.
 389  */
 390 static int copy_strings(int argc, const char __user *const __user *argv,
 391                         struct linux_binprm *bprm)
 392 {
 393         struct page *kmapped_page = NULL;
 394         char *kaddr = NULL;
 395         unsigned long kpos = 0;
 396         int ret;
 397
 398         while (argc-- > 0) {
 399                 const char __user *str;
 400                 int len;
 401                 unsigned long pos;
 402
 403                 if (get_user(str, argv+argc) ||
 404                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 405                         ret = -EFAULT;
 406                         goto out;
 407                 }
 408
 409                 if (!valid_arg_len(bprm, len)) {
 410                         ret = -E2BIG;
 411                         goto out;
 412                 }
 413
 414                 /* We're going to work our way backwords. */
 415                 pos = bprm->p;
 416                 str += len;
 417                 bprm->p -= len;
 418
 419                 while (len > 0) {
 420                         int offset, bytes_to_copy;
 421
 422                         offset = pos % PAGE_SIZE;
 423                         if (offset == 0)
 424                                 offset = PAGE_SIZE;
 425
 426                         bytes_to_copy = offset;
 427                         if (bytes_to_copy > len)
 428                                 bytes_to_copy = len;
 429
 430                         offset -= bytes_to_copy;
 431                         pos -= bytes_to_copy;
 432                         str -= bytes_to_copy;
 433                         len -= bytes_to_copy;
 434
 435                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 436                                 struct page *page;
 437
 438                                 page = get_arg_page(bprm, pos, 1);
 439                                 if (!page) {
 440                                         ret = -E2BIG;
 441                                         goto out;
 442                                 }
 443
 444                                 if (kmapped_page) {
 445                                         flush_kernel_dcache_page(kmapped_page);
 446                                         kunmap(kmapped_page);
 447                                         put_arg_page(kmapped_page);
 448                                 }
 449                                 kmapped_page = page;
 450                                 kaddr = kmap(kmapped_page);
 451                                 kpos = pos & PAGE_MASK;
 452                                 flush_arg_page(bprm, kpos, kmapped_page);
 453                         }
 454                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 455                                 ret = -EFAULT;
 456                                 goto out;
 457                         }
 458                 }
 459         }
 460         ret = 0;
 461 out:
 462         if (kmapped_page) {
 463                 flush_kernel_dcache_page(kmapped_page);
 464                 kunmap(kmapped_page);
 465                 put_arg_page(kmapped_page);
 466         }
 467         return ret;
 468 }
 469
 470 /*
 471  * Like copy_strings, but get argv and its values from kernel memory.
 472  */
 473 int copy_strings_kernel(int argc, const char *const *argv,
 474                         struct linux_binprm *bprm)
 475 {
 476         int r;
 477         mm_segment_t oldfs = get_fs();
 478         set_fs(KERNEL_DS);
 479         r = copy_strings(argc, (const char __user *const  __user *)argv, bprm);
 480         set_fs(oldfs);
 481         return r;
 482 }
 483 EXPORT_SYMBOL(copy_strings_kernel);
 484
 485 #ifdef CONFIG_MMU
 486
 487 /*
 488  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 489  * the binfmt code determines where the new stack should reside, we shift it to
 490  * its final location.  The process proceeds as follows:
 491  *
 492  * 1) Use shift to calculate the new vma endpoints.
 493  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 494  *    arguments passed to subsequent functions are consistent.
 495  * 3) Move vma's page tables to the new range.
 496  * 4) Free up any cleared pgd range.
 497  * 5) Shrink the vma to cover only the new range.
 498  */
 499 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 500 {
 501         struct mm_struct *mm = vma->vm_mm;
 502         unsigned long old_start = vma->vm_start;
 503         unsigned long old_end = vma->vm_end;
 504         unsigned long length = old_end - old_start;
 505         unsigned long new_start = old_start - shift;
 506         unsigned long new_end = old_end - shift;
 507         struct mmu_gather *tlb;
 508
 509         BUG_ON(new_start > new_end);
 510
 511         /*
 512          * ensure there are no vmas between where we want to go
 513          * and where we are
 514          */
 515         if (vma != find_vma(mm, new_start))
 516                 return -EFAULT;
 517
 518         /*
 519          * cover the whole range: [new_start, old_end)
 520          */
 521         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 522                 return -ENOMEM;
 523
 524         /*
 525          * move the page tables downwards, on failure we rely on
 526          * process cleanup to remove whatever mess we made.
 527          */
 528         if (length != move_page_tables(vma, old_start,
 529                                        vma, new_start, length))
 530                 return -ENOMEM;
 531
 532         lru_add_drain();
 533         tlb = tlb_gather_mmu(mm, 0);
 534         if (new_end > old_start) {
 535                 /*
 536                  * when the old and new regions overlap clear from new_end.
 537                  */
 538                 free_pgd_range(tlb, new_end, old_end, new_end,
 539                         vma->vm_next ? vma->vm_next->vm_start : 0);
 540         } else {
 541                 /*
 542                  * otherwise, clean from old_start; this is done to not touch
 543                  * the address space in [new_end, old_start) some architectures
 544                  * have constraints on va-space that make this illegal (IA64) -
 545                  * for the others its just a little faster.
 546                  */
 547                 free_pgd_range(tlb, old_start, old_end, new_end,
 548                         vma->vm_next ? vma->vm_next->vm_start : 0);
 549         }
 550         tlb_finish_mmu(tlb, new_end, old_end);
 551
 552         /*
 553          * Shrink the vma to just the new range.  Always succeeds.
 554          */
 555         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 556
 557         return 0;
 558 }
 559
 560 /*
 561  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 562  * the stack is optionally relocated, and some extra space is added.
 563  */
 564 int setup_arg_pages(struct linux_binprm *bprm,
 565                     unsigned long stack_top,
 566                     int executable_stack)
 567 {
 568         unsigned long ret;
 569         unsigned long stack_shift;
 570         struct mm_struct *mm = current->mm;
 571         struct vm_area_struct *vma = bprm->vma;
 572         struct vm_area_struct *prev = NULL;
 573         unsigned long vm_flags;
 574         unsigned long stack_base;
 575         unsigned long stack_size;
 576         unsigned long stack_expand;
 577         unsigned long rlim_stack;
 578
 579 #ifdef CONFIG_STACK_GROWSUP
 580         /* Limit stack size to 1GB */
 581         stack_base = rlimit_max(RLIMIT_STACK);
 582         if (stack_base > (1 << 30))
 583                 stack_base = 1 << 30;
 584
 585         /* Make sure we didn't let the argument array grow too large. */
 586         if (vma->vm_end - vma->vm_start > stack_base)
 587                 return -ENOMEM;
 588
 589         stack_base = PAGE_ALIGN(stack_top - stack_base);
 590
 591         stack_shift = vma->vm_start - stack_base;
 592         mm->arg_start = bprm->p - stack_shift;
 593         bprm->p = vma->vm_end - stack_shift;
 594 #else
 595         stack_top = arch_align_stack(stack_top);
 596         stack_top = PAGE_ALIGN(stack_top);
 597
 598         if (unlikely(stack_top < mmap_min_addr) ||
 599             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 600                 return -ENOMEM;
 601
 602         stack_shift = vma->vm_end - stack_top;
 603
 604         bprm->p -= stack_shift;
 605         mm->arg_start = bprm->p;
 606 #endif
 607
 608         if (bprm->loader)
 609                 bprm->loader -= stack_shift;
 610         bprm->exec -= stack_shift;
 611
 612         down_write(&mm->mmap_sem);
 613         vm_flags = VM_STACK_FLAGS;
 614
 615         /*
 616          * Adjust stack execute permissions; explicitly enable for
 617          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 618          * (arch default) otherwise.
 619          */
 620         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 621                 vm_flags |= VM_EXEC;
 622         else if (executable_stack == EXSTACK_DISABLE_X)
 623                 vm_flags &= ~VM_EXEC;
 624         vm_flags |= mm->def_flags;
 625         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 626
 627         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 628                         vm_flags);
 629         if (ret)
 630                 goto out_unlock;
 631         BUG_ON(prev != vma);
 632
 633         /* Move stack pages down in memory. */
 634         if (stack_shift) {
 635                 ret = shift_arg_pages(vma, stack_shift);
 636                 if (ret)
 637                         goto out_unlock;
 638         }
 639
 640         /* mprotect_fixup is overkill to remove the temporary stack flags */
 641         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 642
 643         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 644         stack_size = vma->vm_end - vma->vm_start;
 645         /*
 646          * Align this down to a page boundary as expand_stack
 647          * will align it up.
 648          */
 649         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 650 #ifdef CONFIG_STACK_GROWSUP
 651         if (stack_size + stack_expand > rlim_stack)
 652                 stack_base = vma->vm_start + rlim_stack;
 653         else
 654                 stack_base = vma->vm_end + stack_expand;
 655 #else
 656         if (stack_size + stack_expand > rlim_stack)
 657                 stack_base = vma->vm_end - rlim_stack;
 658         else
 659                 stack_base = vma->vm_start - stack_expand;
 660 #endif
 661         current->mm->start_stack = bprm->p;
 662         ret = expand_stack(vma, stack_base);
 663         if (ret)
 664                 ret = -EFAULT;
 665
 666 out_unlock:
 667         up_write(&mm->mmap_sem);
 668         return ret;
 669 }
 670 EXPORT_SYMBOL(setup_arg_pages);
 671
 672 #endif /* CONFIG_MMU */
 673
 674 struct file *open_exec(const char *name)
 675 {
 676         struct file *file;
 677         int err;
 678
 679         file = do_filp_open(AT_FDCWD, name,
 680                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 681                                 MAY_EXEC | MAY_OPEN);
 682         if (IS_ERR(file))
 683                 goto out;
 684
 685         err = -EACCES;
 686         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 687                 goto exit;
 688
 689         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 690                 goto exit;
 691
 692         fsnotify_open(file);
 693
 694         err = deny_write_access(file);
 695         if (err)
 696                 goto exit;
 697
 698 out:
 699         return file;
 700
 701 exit:
 702         fput(file);
 703         return ERR_PTR(err);
 704 }
 705 EXPORT_SYMBOL(open_exec);
 706
 707 int kernel_read(struct file *file, loff_t offset,
 708                 char *addr, unsigned long count)
 709 {
 710         mm_segment_t old_fs;
 711         loff_t pos = offset;
 712         int result;
 713
 714         old_fs = get_fs();
 715         set_fs(get_ds());
 716         /* The cast to a user pointer is valid due to the set_fs() */
 717         result = vfs_read(file, (void __user *)addr, count, &pos);
 718         set_fs(old_fs);
 719         return result;
 720 }
 721
 722 EXPORT_SYMBOL(kernel_read);
 723
 724 static int exec_mmap(struct mm_struct *mm)
 725 {
 726         struct task_struct *tsk;
 727         struct mm_struct * old_mm, *active_mm;
 728
 729         /* Notify parent that we're no longer interested in the old VM */
 730         tsk = current;
 731         old_mm = current->mm;
 732         sync_mm_rss(tsk, old_mm);
 733         mm_release(tsk, old_mm);
 734
 735         if (old_mm) {
 736                 /*
 737                  * Make sure that if there is a core dump in progress
 738                  * for the old mm, we get out and die instead of going
 739                  * through with the exec.  We must hold mmap_sem around
 740                  * checking core_state and changing tsk->mm.
 741                  */
 742                 down_read(&old_mm->mmap_sem);
 743                 if (unlikely(old_mm->core_state)) {
 744                         up_read(&old_mm->mmap_sem);
 745                         return -EINTR;
 746                 }
 747         }
 748         task_lock(tsk);
 749         active_mm = tsk->active_mm;
 750         tsk->mm = mm;
 751         tsk->active_mm = mm;
 752         activate_mm(active_mm, mm);
 753         task_unlock(tsk);
 754         arch_pick_mmap_layout(mm);
 755         if (old_mm) {
 756                 up_read(&old_mm->mmap_sem);
 757                 BUG_ON(active_mm != old_mm);
 758                 mm_update_next_owner(old_mm);
 759                 mmput(old_mm);
 760                 return 0;
 761         }
 762         mmdrop(active_mm);
 763         return 0;
 764 }
 765
 766 /*
 767  * This function makes sure the current process has its own signal table,
 768  * so that flush_signal_handlers can later reset the handlers without
 769  * disturbing other processes.  (Other processes might share the signal
 770  * table via the CLONE_SIGHAND option to clone().)
 771  */
 772 static int de_thread(struct task_struct *tsk)
 773 {
 774         struct signal_struct *sig = tsk->signal;
 775         struct sighand_struct *oldsighand = tsk->sighand;
 776         spinlock_t *lock = &oldsighand->siglock;
 777
 778         if (thread_group_empty(tsk))
 779                 goto no_thread_group;
 780
 781         /*
 782          * Kill all other threads in the thread group.
 783          */
 784         spin_lock_irq(lock);
 785         if (signal_group_exit(sig)) {
 786                 /*
 787                  * Another group action in progress, just
 788                  * return so that the signal is processed.
 789                  */
 790                 spin_unlock_irq(lock);
 791                 return -EAGAIN;
 792         }
 793
 794         sig->group_exit_task = tsk;
 795         sig->notify_count = zap_other_threads(tsk);
 796         if (!thread_group_leader(tsk))
 797                 sig->notify_count--;
 798
 799         while (sig->notify_count) {
 800                 __set_current_state(TASK_UNINTERRUPTIBLE);
 801                 spin_unlock_irq(lock);
 802                 schedule();
 803                 spin_lock_irq(lock);
 804         }
 805         spin_unlock_irq(lock);
 806
 807         /*
 808          * At this point all other threads have exited, all we have to
 809          * do is to wait for the thread group leader to become inactive,
 810          * and to assume its PID:
 811          */
 812         if (!thread_group_leader(tsk)) {
 813                 struct task_struct *leader = tsk->group_leader;
 814
 815                 sig->notify_count = -1; /* for exit_notify() */
 816                 for (;;) {
 817                         write_lock_irq(&tasklist_lock);
 818                         if (likely(leader->exit_state))
 819                                 break;
 820                         __set_current_state(TASK_UNINTERRUPTIBLE);
 821                         write_unlock_irq(&tasklist_lock);
 822                         schedule();
 823                 }
 824
 825                 /*
 826                  * The only record we have of the real-time age of a
 827                  * process, regardless of execs it's done, is start_time.
 828                  * All the past CPU time is accumulated in signal_struct
 829                  * from sister threads now dead.  But in this non-leader
 830                  * exec, nothing survives from the original leader thread,
 831                  * whose birth marks the true age of this process now.
 832                  * When we take on its identity by switching to its PID, we
 833                  * also take its birthdate (always earlier than our own).
 834                  */
 835                 tsk->start_time = leader->start_time;
 836
 837                 BUG_ON(!same_thread_group(leader, tsk));
 838                 BUG_ON(has_group_leader_pid(tsk));
 839                 /*
 840                  * An exec() starts a new thread group with the
 841                  * TGID of the previous thread group. Rehash the
 842                  * two threads with a switched PID, and release
 843                  * the former thread group leader:
 844                  */
 845
 846                 /* Become a process group leader with the old leader's pid.
 847                  * The old leader becomes a thread of the this thread group.
 848                  * Note: The old leader also uses this pid until release_task
 849                  *       is called.  Odd but simple and correct.
 850                  */
 851                 detach_pid(tsk, PIDTYPE_PID);
 852                 tsk->pid = leader->pid;
 853                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 854                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 855                 transfer_pid(leader, tsk, PIDTYPE_SID);
 856
 857                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 858                 list_replace_init(&leader->sibling, &tsk->sibling);
 859
 860                 tsk->group_leader = tsk;
 861                 leader->group_leader = tsk;
 862
 863                 tsk->exit_signal = SIGCHLD;
 864
 865                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 866                 leader->exit_state = EXIT_DEAD;
 867                 write_unlock_irq(&tasklist_lock);
 868
 869                 release_task(leader);
 870         }
 871
 872         sig->group_exit_task = NULL;
 873         sig->notify_count = 0;
 874
 875 no_thread_group:
 876         if (current->mm)
 877                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 878
 879         exit_itimers(sig);
 880         flush_itimer_signals();
 881
 882         if (atomic_read(&oldsighand->count) != 1) {
 883                 struct sighand_struct *newsighand;
 884                 /*
 885                  * This ->sighand is shared with the CLONE_SIGHAND
 886                  * but not CLONE_THREAD task, switch to the new one.
 887                  */
 888                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 889                 if (!newsighand)
 890                         return -ENOMEM;
 891
 892                 atomic_set(&newsighand->count, 1);
 893                 memcpy(newsighand->action, oldsighand->action,
 894                        sizeof(newsighand->action));
 895
 896                 write_lock_irq(&tasklist_lock);
 897                 spin_lock(&oldsighand->siglock);
 898                 rcu_assign_pointer(tsk->sighand, newsighand);
 899                 spin_unlock(&oldsighand->siglock);
 900                 write_unlock_irq(&tasklist_lock);
 901
 902                 __cleanup_sighand(oldsighand);
 903         }
 904
 905         BUG_ON(!thread_group_leader(tsk));
 906         return 0;
 907 }
 908
 909 /*
 910  * These functions flushes out all traces of the currently running executable
 911  * so that a new one can be started
 912  */
 913 static void flush_old_files(struct files_struct * files)
 914 {
 915         long j = -1;
 916         struct fdtable *fdt;
 917
 918         spin_lock(&files->file_lock);
 919         for (;;) {
 920                 unsigned long set, i;
 921
 922                 j++;
 923                 i = j * __NFDBITS;
 924                 fdt = files_fdtable(files);
 925                 if (i >= fdt->max_fds)
 926                         break;
 927                 set = fdt->close_on_exec->fds_bits[j];
 928                 if (!set)
 929                         continue;
 930                 fdt->close_on_exec->fds_bits[j] = 0;
 931                 spin_unlock(&files->file_lock);
 932                 for ( ; set ; i++,set >>= 1) {
 933                         if (set & 1) {
 934                                 sys_close(i);
 935                         }
 936                 }
 937                 spin_lock(&files->file_lock);
 938
 939         }
 940         spin_unlock(&files->file_lock);
 941 }
 942
 943 char *get_task_comm(char *buf, struct task_struct *tsk)
 944 {
 945         /* buf must be at least sizeof(tsk->comm) in size */
 946         task_lock(tsk);
 947         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 948         task_unlock(tsk);
 949         return buf;
 950 }
 951
 952 void set_task_comm(struct task_struct *tsk, char *buf)
 953 {
 954         task_lock(tsk);
 955
 956         /*
 957          * Threads may access current->comm without holding
 958          * the task lock, so write the string carefully.
 959          * Readers without a lock may see incomplete new
 960          * names but are safe from non-terminating string reads.
 961          */
 962         memset(tsk->comm, 0, TASK_COMM_LEN);
 963         wmb();
 964         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 965         task_unlock(tsk);
 966         perf_event_comm(tsk);
 967 }
 968
 969 int flush_old_exec(struct linux_binprm * bprm)
 970 {
 971         int retval;
 972
 973         /*
 974          * Make sure we have a private signal table and that
 975          * we are unassociated from the previous thread group.
 976          */
 977         retval = de_thread(current);
 978         if (retval)
 979                 goto out;
 980
 981         set_mm_exe_file(bprm->mm, bprm->file);
 982
 983         /*
 984          * Release all of the old mmap stuff
 985          */
 986         retval = exec_mmap(bprm->mm);
 987         if (retval)
 988                 goto out;
 989
 990         bprm->mm = NULL;                /* We're using it now */
 991
 992         current->flags &= ~PF_RANDOMIZE;
 993         flush_thread();
 994         current->personality &= ~bprm->per_clear;
 995
 996         return 0;
 997
 998 out:
 999         return retval;
1000 }
1001 EXPORT_SYMBOL(flush_old_exec);
1002
1003 void setup_new_exec(struct linux_binprm * bprm)
1004 {
1005         int i, ch;
1006         const char *name;
1007         char tcomm[sizeof(current->comm)];
1008
1009         arch_pick_mmap_layout(current->mm);
1010
1011         /* This is the point of no return */
1012         current->sas_ss_sp = current->sas_ss_size = 0;
1013
1014         if (current_euid() == current_uid() && current_egid() == current_gid())
1015                 set_dumpable(current->mm, 1);
1016         else
1017                 set_dumpable(current->mm, suid_dumpable);
1018
1019         name = bprm->filename;
1020
1021         /* Copies the binary name from after last slash */
1022         for (i=0; (ch = *(name++)) != '\0';) {
1023                 if (ch == '/')
1024                         i = 0; /* overwrite what we wrote */
1025                 else
1026                         if (i < (sizeof(tcomm) - 1))
1027                                 tcomm[i++] = ch;
1028         }
1029         tcomm[i] = '\0';
1030         set_task_comm(current, tcomm);
1031
1032         /* Set the new mm task size. We have to do that late because it may
1033          * depend on TIF_32BIT which is only updated in flush_thread() on
1034          * some architectures like powerpc
1035          */
1036         current->mm->task_size = TASK_SIZE;
1037
1038         /* install the new credentials */
1039         if (bprm->cred->uid != current_euid() ||
1040             bprm->cred->gid != current_egid()) {
1041                 current->pdeath_signal = 0;
1042         } else if (file_permission(bprm->file, MAY_READ) ||
1043                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1044                 set_dumpable(current->mm, suid_dumpable);
1045         }
1046
1047         /*
1048          * Flush performance counters when crossing a
1049          * security domain:
1050          */
1051         if (!get_dumpable(current->mm))
1052                 perf_event_exit_task(current);
1053
1054         /* An exec changes our domain. We are no longer part of the thread
1055            group */
1056
1057         current->self_exec_id++;
1058
1059         flush_signal_handlers(current, 0);
1060         flush_old_files(current->files);
1061 }
1062 EXPORT_SYMBOL(setup_new_exec);
1063
1064 /*
1065  * Prepare credentials and lock ->cred_guard_mutex.
1066  * install_exec_creds() commits the new creds and drops the lock.
1067  * Or, if exec fails before, free_bprm() should release ->cred and
1068  * and unlock.
1069  */
1070 int prepare_bprm_creds(struct linux_binprm *bprm)
1071 {
1072         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1073                 return -ERESTARTNOINTR;
1074
1075         bprm->cred = prepare_exec_creds();
1076         if (likely(bprm->cred))
1077                 return 0;
1078
1079         mutex_unlock(&current->cred_guard_mutex);
1080         return -ENOMEM;
1081 }
1082
1083 void free_bprm(struct linux_binprm *bprm)
1084 {
1085         free_arg_pages(bprm);
1086         if (bprm->cred) {
1087                 mutex_unlock(&current->cred_guard_mutex);
1088                 abort_creds(bprm->cred);
1089         }
1090         kfree(bprm);
1091 }
1092
1093 /*
1094  * install the new credentials for this executable
1095  */
1096 void install_exec_creds(struct linux_binprm *bprm)
1097 {
1098         security_bprm_committing_creds(bprm);
1099
1100         commit_creds(bprm->cred);
1101         bprm->cred = NULL;
1102         /*
1103          * cred_guard_mutex must be held at least to this point to prevent
1104          * ptrace_attach() from altering our determination of the task's
1105          * credentials; any time after this it may be unlocked.
1106          */
1107         security_bprm_committed_creds(bprm);
1108         mutex_unlock(&current->cred_guard_mutex);
1109 }
1110 EXPORT_SYMBOL(install_exec_creds);
1111
1112 /*
1113  * determine how safe it is to execute the proposed program
1114  * - the caller must hold current->cred_guard_mutex to protect against
1115  *   PTRACE_ATTACH
1116  */
1117 int check_unsafe_exec(struct linux_binprm *bprm)
1118 {
1119         struct task_struct *p = current, *t;
1120         unsigned n_fs;
1121         int res = 0;
1122
1123         bprm->unsafe = tracehook_unsafe_exec(p);
1124
1125         n_fs = 1;
1126         spin_lock(&p->fs->lock);
1127         rcu_read_lock();
1128         for (t = next_thread(p); t != p; t = next_thread(t)) {
1129                 if (t->fs == p->fs)
1130                         n_fs++;
1131         }
1132         rcu_read_unlock();
1133
1134         if (p->fs->users > n_fs) {
1135                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1136         } else {
1137                 res = -EAGAIN;
1138                 if (!p->fs->in_exec) {
1139                         p->fs->in_exec = 1;
1140                         res = 1;
1141                 }
1142         }
1143         spin_unlock(&p->fs->lock);
1144
1145         return res;
1146 }
1147
1148 /*
1149  * Fill the binprm structure from the inode.
1150  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1151  *
1152  * This may be called multiple times for binary chains (scripts for example).
1153  */
1154 int prepare_binprm(struct linux_binprm *bprm)
1155 {
1156         umode_t mode;
1157         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1158         int retval;
1159
1160         mode = inode->i_mode;
1161         if (bprm->file->f_op == NULL)
1162                 return -EACCES;
1163
1164         /* clear any previous set[ug]id data from a previous binary */
1165         bprm->cred->euid = current_euid();
1166         bprm->cred->egid = current_egid();
1167
1168         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1169                 /* Set-uid? */
1170                 if (mode & S_ISUID) {
1171                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1172                         bprm->cred->euid = inode->i_uid;
1173                 }
1174
1175                 /* Set-gid? */
1176                 /*
1177                  * If setgid is set but no group execute bit then this
1178                  * is a candidate for mandatory locking, not a setgid
1179                  * executable.
1180                  */
1181                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1182                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1183                         bprm->cred->egid = inode->i_gid;
1184                 }
1185         }
1186
1187         /* fill in binprm security blob */
1188         retval = security_bprm_set_creds(bprm);
1189         if (retval)
1190                 return retval;
1191         bprm->cred_prepared = 1;
1192
1193         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1194         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1195 }
1196
1197 EXPORT_SYMBOL(prepare_binprm);
1198
1199 /*
1200  * Arguments are '\0' separated strings found at the location bprm->p
1201  * points to; chop off the first by relocating brpm->p to right after
1202  * the first '\0' encountered.
1203  */
1204 int remove_arg_zero(struct linux_binprm *bprm)
1205 {
1206         int ret = 0;
1207         unsigned long offset;
1208         char *kaddr;
1209         struct page *page;
1210
1211         if (!bprm->argc)
1212                 return 0;
1213
1214         do {
1215                 offset = bprm->p & ~PAGE_MASK;
1216                 page = get_arg_page(bprm, bprm->p, 0);
1217                 if (!page) {
1218                         ret = -EFAULT;
1219                         goto out;
1220                 }
1221                 kaddr = kmap_atomic(page, KM_USER0);
1222
1223                 for (; offset < PAGE_SIZE && kaddr[offset];
1224                                 offset++, bprm->p++)
1225                         ;
1226
1227                 kunmap_atomic(kaddr, KM_USER0);
1228                 put_arg_page(page);
1229
1230                 if (offset == PAGE_SIZE)
1231                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1232         } while (offset == PAGE_SIZE);
1233
1234         bprm->p++;
1235         bprm->argc--;
1236         ret = 0;
1237
1238 out:
1239         return ret;
1240 }
1241 EXPORT_SYMBOL(remove_arg_zero);
1242
1243 /*
1244  * cycle the list of binary formats handler, until one recognizes the image
1245  */
1246 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1247 {
1248         unsigned int depth = bprm->recursion_depth;
1249         int try,retval;
1250         struct linux_binfmt *fmt;
1251
1252         retval = security_bprm_check(bprm);
1253         if (retval)
1254                 return retval;
1255
1256         /* kernel module loader fixup */
1257         /* so we don't try to load run modprobe in kernel space. */
1258         set_fs(USER_DS);
1259
1260         retval = audit_bprm(bprm);
1261         if (retval)
1262                 return retval;
1263
1264         retval = -ENOENT;
1265         for (try=0; try<2; try++) {
1266                 read_lock(&binfmt_lock);
1267                 list_for_each_entry(fmt, &formats, lh) {
1268                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1269                         if (!fn)
1270                                 continue;
1271                         if (!try_module_get(fmt->module))
1272                                 continue;
1273                         read_unlock(&binfmt_lock);
1274                         retval = fn(bprm, regs);
1275                         /*
1276                          * Restore the depth counter to its starting value
1277                          * in this call, so we don't have to rely on every
1278                          * load_binary function to restore it on return.
1279                          */
1280                         bprm->recursion_depth = depth;
1281                         if (retval >= 0) {
1282                                 if (depth == 0)
1283                                         tracehook_report_exec(fmt, bprm, regs);
1284                                 put_binfmt(fmt);
1285                                 allow_write_access(bprm->file);
1286                                 if (bprm->file)
1287                                         fput(bprm->file);
1288                                 bprm->file = NULL;
1289                                 current->did_exec = 1;
1290                                 proc_exec_connector(current);
1291                                 return retval;
1292                         }
1293                         read_lock(&binfmt_lock);
1294                         put_binfmt(fmt);
1295                         if (retval != -ENOEXEC || bprm->mm == NULL)
1296                                 break;
1297                         if (!bprm->file) {
1298                                 read_unlock(&binfmt_lock);
1299                                 return retval;
1300                         }
1301                 }
1302                 read_unlock(&binfmt_lock);
1303                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1304                         break;
1305 #ifdef CONFIG_MODULES
1306                 } else {
1307 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1308                         if (printable(bprm->buf[0]) &&
1309                             printable(bprm->buf[1]) &&
1310                             printable(bprm->buf[2]) &&
1311                             printable(bprm->buf[3]))
1312                                 break; /* -ENOEXEC */
1313                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1314 #endif
1315                 }
1316         }
1317         return retval;
1318 }
1319
1320 EXPORT_SYMBOL(search_binary_handler);
1321
1322 /*
1323  * sys_execve() executes a new program.
1324  */
1325 int do_execve(const char * filename,
1326         const char __user *const __user *argv,
1327         const char __user *const __user *envp,
1328         struct pt_regs * regs)
1329 {
1330         struct linux_binprm *bprm;
1331         struct file *file;
1332         struct files_struct *displaced;
1333         bool clear_in_exec;
1334         int retval;
1335
1336         retval = unshare_files(&displaced);
1337         if (retval)
1338                 goto out_ret;
1339
1340         retval = -ENOMEM;
1341         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1342         if (!bprm)
1343                 goto out_files;
1344
1345         retval = prepare_bprm_creds(bprm);
1346         if (retval)
1347                 goto out_free;
1348
1349         retval = check_unsafe_exec(bprm);
1350         if (retval < 0)
1351                 goto out_free;
1352         clear_in_exec = retval;
1353         current->in_execve = 1;
1354
1355         file = open_exec(filename);
1356         retval = PTR_ERR(file);
1357         if (IS_ERR(file))
1358                 goto out_unmark;
1359
1360         sched_exec();
1361
1362         bprm->file = file;
1363         bprm->filename = filename;
1364         bprm->interp = filename;
1365
1366         retval = bprm_mm_init(bprm);
1367         if (retval)
1368                 goto out_file;
1369
1370         bprm->argc = count(argv, MAX_ARG_STRINGS);
1371         if ((retval = bprm->argc) < 0)
1372                 goto out;
1373
1374         bprm->envc = count(envp, MAX_ARG_STRINGS);
1375         if ((retval = bprm->envc) < 0)
1376                 goto out;
1377
1378         retval = prepare_binprm(bprm);
1379         if (retval < 0)
1380                 goto out;
1381
1382         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1383         if (retval < 0)
1384                 goto out;
1385
1386         bprm->exec = bprm->p;
1387         retval = copy_strings(bprm->envc, envp, bprm);
1388         if (retval < 0)
1389                 goto out;
1390
1391         retval = copy_strings(bprm->argc, argv, bprm);
1392         if (retval < 0)
1393                 goto out;
1394
1395         current->flags &= ~PF_KTHREAD;
1396         retval = search_binary_handler(bprm,regs);
1397         if (retval < 0)
1398                 goto out;
1399
1400         /* execve succeeded */
1401         current->fs->in_exec = 0;
1402         current->in_execve = 0;
1403         acct_update_integrals(current);
1404         free_bprm(bprm);
1405         if (displaced)
1406                 put_files_struct(displaced);
1407         return retval;
1408
1409 out:
1410         if (bprm->mm)
1411                 mmput (bprm->mm);
1412
1413 out_file:
1414         if (bprm->file) {
1415                 allow_write_access(bprm->file);
1416                 fput(bprm->file);
1417         }
1418
1419 out_unmark:
1420         if (clear_in_exec)
1421                 current->fs->in_exec = 0;
1422         current->in_execve = 0;
1423
1424 out_free:
1425         free_bprm(bprm);
1426
1427 out_files:
1428         if (displaced)
1429                 reset_files_struct(displaced);
1430 out_ret:
1431         return retval;
1432 }
1433
1434 void set_binfmt(struct linux_binfmt *new)
1435 {
1436         struct mm_struct *mm = current->mm;
1437
1438         if (mm->binfmt)
1439                 module_put(mm->binfmt->module);
1440
1441         mm->binfmt = new;
1442         if (new)
1443                 __module_get(new->module);
1444 }
1445
1446 EXPORT_SYMBOL(set_binfmt);
1447
1448 /* format_corename will inspect the pattern parameter, and output a
1449  * name into corename, which must have space for at least
1450  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1451  */
1452 static int format_corename(char *corename, long signr)
1453 {
1454         const struct cred *cred = current_cred();
1455         const char *pat_ptr = core_pattern;
1456         int ispipe = (*pat_ptr == '|');
1457         char *out_ptr = corename;
1458         char *const out_end = corename + CORENAME_MAX_SIZE;
1459         int rc;
1460         int pid_in_pattern = 0;
1461
1462         /* Repeat as long as we have more pattern to process and more output
1463            space */
1464         while (*pat_ptr) {
1465                 if (*pat_ptr != '%') {
1466                         if (out_ptr == out_end)
1467                                 goto out;
1468                         *out_ptr++ = *pat_ptr++;
1469                 } else {
1470                         switch (*++pat_ptr) {
1471                         case 0:
1472                                 goto out;
1473                         /* Double percent, output one percent */
1474                         case '%':
1475                                 if (out_ptr == out_end)
1476                                         goto out;
1477                                 *out_ptr++ = '%';
1478                                 break;
1479                         /* pid */
1480                         case 'p':
1481                                 pid_in_pattern = 1;
1482                                 rc = snprintf(out_ptr, out_end - out_ptr,
1483                                               "%d", task_tgid_vnr(current));
1484                                 if (rc > out_end - out_ptr)
1485                                         goto out;
1486                                 out_ptr += rc;
1487                                 break;
1488                         /* uid */
1489                         case 'u':
1490                                 rc = snprintf(out_ptr, out_end - out_ptr,
1491                                               "%d", cred->uid);
1492                                 if (rc > out_end - out_ptr)
1493                                         goto out;
1494                                 out_ptr += rc;
1495                                 break;
1496                         /* gid */
1497                         case 'g':
1498                                 rc = snprintf(out_ptr, out_end - out_ptr,
1499                                               "%d", cred->gid);
1500                                 if (rc > out_end - out_ptr)
1501                                         goto out;
1502                                 out_ptr += rc;
1503                                 break;
1504                         /* signal that caused the coredump */
1505                         case 's':
1506                                 rc = snprintf(out_ptr, out_end - out_ptr,
1507                                               "%ld", signr);
1508                                 if (rc > out_end - out_ptr)
1509                                         goto out;
1510                                 out_ptr += rc;
1511                                 break;
1512                         /* UNIX time of coredump */
1513                         case 't': {
1514                                 struct timeval tv;
1515                                 do_gettimeofday(&tv);
1516                                 rc = snprintf(out_ptr, out_end - out_ptr,
1517                                               "%lu", tv.tv_sec);
1518                                 if (rc > out_end - out_ptr)
1519                                         goto out;
1520                                 out_ptr += rc;
1521                                 break;
1522                         }
1523                         /* hostname */
1524                         case 'h':
1525                                 down_read(&uts_sem);
1526                                 rc = snprintf(out_ptr, out_end - out_ptr,
1527                                               "%s", utsname()->nodename);
1528                                 up_read(&uts_sem);
1529                                 if (rc > out_end - out_ptr)
1530                                         goto out;
1531                                 out_ptr += rc;
1532                                 break;
1533                         /* executable */
1534                         case 'e':
1535                                 rc = snprintf(out_ptr, out_end - out_ptr,
1536                                               "%s", current->comm);
1537                                 if (rc > out_end - out_ptr)
1538                                         goto out;
1539                                 out_ptr += rc;
1540                                 break;
1541                         /* core limit size */
1542                         case 'c':
1543                                 rc = snprintf(out_ptr, out_end - out_ptr,
1544                                               "%lu", rlimit(RLIMIT_CORE));
1545                                 if (rc > out_end - out_ptr)
1546                                         goto out;
1547                                 out_ptr += rc;
1548                                 break;
1549                         default:
1550                                 break;
1551                         }
1552                         ++pat_ptr;
1553                 }
1554         }
1555         /* Backward compatibility with core_uses_pid:
1556          *
1557          * If core_pattern does not include a %p (as is the default)
1558          * and core_uses_pid is set, then .%pid will be appended to
1559          * the filename. Do not do this for piped commands. */
1560         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1561                 rc = snprintf(out_ptr, out_end - out_ptr,
1562                               ".%d", task_tgid_vnr(current));
1563                 if (rc > out_end - out_ptr)
1564                         goto out;
1565                 out_ptr += rc;
1566         }
1567 out:
1568         *out_ptr = 0;
1569         return ispipe;
1570 }
1571
1572 static int zap_process(struct task_struct *start, int exit_code)
1573 {
1574         struct task_struct *t;
1575         int nr = 0;
1576
1577         start->signal->flags = SIGNAL_GROUP_EXIT;
1578         start->signal->group_exit_code = exit_code;
1579         start->signal->group_stop_count = 0;
1580
1581         t = start;
1582         do {
1583                 if (t != current && t->mm) {
1584                         sigaddset(&t->pending.signal, SIGKILL);
1585                         signal_wake_up(t, 1);
1586                         nr++;
1587                 }
1588         } while_each_thread(start, t);
1589
1590         return nr;
1591 }
1592
1593 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1594                                 struct core_state *core_state, int exit_code)
1595 {
1596         struct task_struct *g, *p;
1597         unsigned long flags;
1598         int nr = -EAGAIN;
1599
1600         spin_lock_irq(&tsk->sighand->siglock);
1601         if (!signal_group_exit(tsk->signal)) {
1602                 mm->core_state = core_state;
1603                 nr = zap_process(tsk, exit_code);
1604         }
1605         spin_unlock_irq(&tsk->sighand->siglock);
1606         if (unlikely(nr < 0))
1607                 return nr;
1608
1609         if (atomic_read(&mm->mm_users) == nr + 1)
1610                 goto done;
1611         /*
1612          * We should find and kill all tasks which use this mm, and we should
1613          * count them correctly into ->nr_threads. We don't take tasklist
1614          * lock, but this is safe wrt:
1615          *
1616          * fork:
1617          *      None of sub-threads can fork after zap_process(leader). All
1618          *      processes which were created before this point should be
1619          *      visible to zap_threads() because copy_process() adds the new
1620          *      process to the tail of init_task.tasks list, and lock/unlock
1621          *      of ->siglock provides a memory barrier.
1622          *
1623          * do_exit:
1624          *      The caller holds mm->mmap_sem. This means that the task which
1625          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1626          *      its ->mm.
1627          *
1628          * de_thread:
1629          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1630          *      we must see either old or new leader, this does not matter.
1631          *      However, it can change p->sighand, so lock_task_sighand(p)
1632          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1633          *      it can't fail.
1634          *
1635          *      Note also that "g" can be the old leader with ->mm == NULL
1636          *      and already unhashed and thus removed from ->thread_group.
1637          *      This is OK, __unhash_process()->list_del_rcu() does not
1638          *      clear the ->next pointer, we will find the new leader via
1639          *      next_thread().
1640          */
1641         rcu_read_lock();
1642         for_each_process(g) {
1643                 if (g == tsk->group_leader)
1644                         continue;
1645                 if (g->flags & PF_KTHREAD)
1646                         continue;
1647                 p = g;
1648                 do {
1649                         if (p->mm) {
1650                                 if (unlikely(p->mm == mm)) {
1651                                         lock_task_sighand(p, &flags);
1652                                         nr += zap_process(p, exit_code);
1653                                         unlock_task_sighand(p, &flags);
1654                                 }
1655                                 break;
1656                         }
1657                 } while_each_thread(g, p);
1658         }
1659         rcu_read_unlock();
1660 done:
1661         atomic_set(&core_state->nr_threads, nr);
1662         return nr;
1663 }
1664
1665 static int coredump_wait(int exit_code, struct core_state *core_state)
1666 {
1667         struct task_struct *tsk = current;
1668         struct mm_struct *mm = tsk->mm;
1669         struct completion *vfork_done;
1670         int core_waiters = -EBUSY;
1671
1672         init_completion(&core_state->startup);
1673         core_state->dumper.task = tsk;
1674         core_state->dumper.next = NULL;
1675
1676         down_write(&mm->mmap_sem);
1677         if (!mm->core_state)
1678                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1679         up_write(&mm->mmap_sem);
1680
1681         if (unlikely(core_waiters < 0))
1682                 goto fail;
1683
1684         /*
1685          * Make sure nobody is waiting for us to release the VM,
1686          * otherwise we can deadlock when we wait on each other
1687          */
1688         vfork_done = tsk->vfork_done;
1689         if (vfork_done) {
1690                 tsk->vfork_done = NULL;
1691                 complete(vfork_done);
1692         }
1693
1694         if (core_waiters)
1695                 wait_for_completion(&core_state->startup);
1696 fail:
1697         return core_waiters;
1698 }
1699
1700 static void coredump_finish(struct mm_struct *mm)
1701 {
1702         struct core_thread *curr, *next;
1703         struct task_struct *task;
1704
1705         next = mm->core_state->dumper.next;
1706         while ((curr = next) != NULL) {
1707                 next = curr->next;
1708                 task = curr->task;
1709                 /*
1710                  * see exit_mm(), curr->task must not see
1711                  * ->task == NULL before we read ->next.
1712                  */
1713                 smp_mb();
1714                 curr->task = NULL;
1715                 wake_up_process(task);
1716         }
1717
1718         mm->core_state = NULL;
1719 }
1720
1721 /*
1722  * set_dumpable converts traditional three-value dumpable to two flags and
1723  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1724  * these bits are not changed atomically.  So get_dumpable can observe the
1725  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1726  * return either old dumpable or new one by paying attention to the order of
1727  * modifying the bits.
1728  *
1729  * dumpable |   mm->flags (binary)
1730  * old  new | initial interim  final
1731  * ---------+-----------------------
1732  *  0    1  |   00      01      01
1733  *  0    2  |   00      10(*)   11
1734  *  1    0  |   01      00      00
1735  *  1    2  |   01      11      11
1736  *  2    0  |   11      10(*)   00
1737  *  2    1  |   11      11      01
1738  *
1739  * (*) get_dumpable regards interim value of 10 as 11.
1740  */
1741 void set_dumpable(struct mm_struct *mm, int value)
1742 {
1743         switch (value) {
1744         case 0:
1745                 clear_bit(MMF_DUMPABLE, &mm->flags);
1746                 smp_wmb();
1747                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1748                 break;
1749         case 1:
1750                 set_bit(MMF_DUMPABLE, &mm->flags);
1751                 smp_wmb();
1752                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1753                 break;
1754         case 2:
1755                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1756                 smp_wmb();
1757                 set_bit(MMF_DUMPABLE, &mm->flags);
1758                 break;
1759         }
1760 }
1761
1762 static int __get_dumpable(unsigned long mm_flags)
1763 {
1764         int ret;
1765
1766         ret = mm_flags & MMF_DUMPABLE_MASK;
1767         return (ret >= 2) ? 2 : ret;
1768 }
1769
1770 int get_dumpable(struct mm_struct *mm)
1771 {
1772         return __get_dumpable(mm->flags);
1773 }
1774
1775 static void wait_for_dump_helpers(struct file *file)
1776 {
1777         struct pipe_inode_info *pipe;
1778
1779         pipe = file->f_path.dentry->d_inode->i_pipe;
1780
1781         pipe_lock(pipe);
1782         pipe->readers++;
1783         pipe->writers--;
1784
1785         while ((pipe->readers > 1) && (!signal_pending(current))) {
1786                 wake_up_interruptible_sync(&pipe->wait);
1787                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1788                 pipe_wait(pipe);
1789         }
1790
1791         pipe->readers--;
1792         pipe->writers++;
1793         pipe_unlock(pipe);
1794
1795 }
1796
1797
1798 /*
1799  * uhm_pipe_setup
1800  * helper function to customize the process used
1801  * to collect the core in userspace.  Specifically
1802  * it sets up a pipe and installs it as fd 0 (stdin)
1803  * for the process.  Returns 0 on success, or
1804  * PTR_ERR on failure.
1805  * Note that it also sets the core limit to 1.  This
1806  * is a special value that we use to trap recursive
1807  * core dumps
1808  */
1809 static int umh_pipe_setup(struct subprocess_info *info)
1810 {
1811         struct file *rp, *wp;
1812         struct fdtable *fdt;
1813         struct coredump_params *cp = (struct coredump_params *)info->data;
1814         struct files_struct *cf = current->files;
1815
1816         wp = create_write_pipe(0);
1817         if (IS_ERR(wp))
1818                 return PTR_ERR(wp);
1819
1820         rp = create_read_pipe(wp, 0);
1821         if (IS_ERR(rp)) {
1822                 free_write_pipe(wp);
1823                 return PTR_ERR(rp);
1824         }
1825
1826         cp->file = wp;
1827
1828         sys_close(0);
1829         fd_install(0, rp);
1830         spin_lock(&cf->file_lock);
1831         fdt = files_fdtable(cf);
1832         FD_SET(0, fdt->open_fds);
1833         FD_CLR(0, fdt->close_on_exec);
1834         spin_unlock(&cf->file_lock);
1835
1836         /* and disallow core files too */
1837         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1838
1839         return 0;
1840 }
1841
1842 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1843 {
1844         struct core_state core_state;
1845         char corename[CORENAME_MAX_SIZE + 1];
1846         struct mm_struct *mm = current->mm;
1847         struct linux_binfmt * binfmt;
1848         const struct cred *old_cred;
1849         struct cred *cred;
1850         int retval = 0;
1851         int flag = 0;
1852         int ispipe;
1853         static atomic_t core_dump_count = ATOMIC_INIT(0);
1854         struct coredump_params cprm = {
1855                 .signr = signr,
1856                 .regs = regs,
1857                 .limit = rlimit(RLIMIT_CORE),
1858                 /*
1859                  * We must use the same mm->flags while dumping core to avoid
1860                  * inconsistency of bit flags, since this flag is not protected
1861                  * by any locks.
1862                  */
1863                 .mm_flags = mm->flags,
1864         };
1865
1866         audit_core_dumps(signr);
1867
1868         binfmt = mm->binfmt;
1869         if (!binfmt || !binfmt->core_dump)
1870                 goto fail;
1871         if (!__get_dumpable(cprm.mm_flags))
1872                 goto fail;
1873
1874         cred = prepare_creds();
1875         if (!cred)
1876                 goto fail;
1877         /*
1878          *      We cannot trust fsuid as being the "true" uid of the
1879          *      process nor do we know its entire history. We only know it
1880          *      was tainted so we dump it as root in mode 2.
1881          */
1882         if (__get_dumpable(cprm.mm_flags) == 2) {
1883                 /* Setuid core dump mode */
1884                 flag = O_EXCL;          /* Stop rewrite attacks */
1885                 cred->fsuid = 0;        /* Dump root private */
1886         }
1887
1888         retval = coredump_wait(exit_code, &core_state);
1889         if (retval < 0)
1890                 goto fail_creds;
1891
1892         old_cred = override_creds(cred);
1893
1894         /*
1895          * Clear any false indication of pending signals that might
1896          * be seen by the filesystem code called to write the core file.
1897          */
1898         clear_thread_flag(TIF_SIGPENDING);
1899
1900         ispipe = format_corename(corename, signr);
1901
1902         if (ispipe) {
1903                 int dump_count;
1904                 char **helper_argv;
1905
1906                 if (cprm.limit == 1) {
1907                         /*
1908                          * Normally core limits are irrelevant to pipes, since
1909                          * we're not writing to the file system, but we use
1910                          * cprm.limit of 1 here as a speacial value. Any
1911                          * non-1 limit gets set to RLIM_INFINITY below, but
1912                          * a limit of 0 skips the dump.  This is a consistent
1913                          * way to catch recursive crashes.  We can still crash
1914                          * if the core_pattern binary sets RLIM_CORE =  !1
1915                          * but it runs as root, and can do lots of stupid things
1916                          * Note that we use task_tgid_vnr here to grab the pid
1917                          * of the process group leader.  That way we get the
1918                          * right pid if a thread in a multi-threaded
1919                          * core_pattern process dies.
1920                          */
1921                         printk(KERN_WARNING
1922                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
1923                                 task_tgid_vnr(current), current->comm);
1924                         printk(KERN_WARNING "Aborting core\n");
1925                         goto fail_unlock;
1926                 }
1927                 cprm.limit = RLIM_INFINITY;
1928
1929                 dump_count = atomic_inc_return(&core_dump_count);
1930                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1931                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1932                                task_tgid_vnr(current), current->comm);
1933                         printk(KERN_WARNING "Skipping core dump\n");
1934                         goto fail_dropcount;
1935                 }
1936
1937                 helper_argv = argv_split(GFP_KERNEL, corename+1, NULL);
1938                 if (!helper_argv) {
1939                         printk(KERN_WARNING "%s failed to allocate memory\n",
1940                                __func__);
1941                         goto fail_dropcount;
1942                 }
1943
1944                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
1945                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
1946                                         NULL, &cprm);
1947                 argv_free(helper_argv);
1948                 if (retval) {
1949                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1950                                corename);
1951                         goto close_fail;
1952                 }
1953         } else {
1954                 struct inode *inode;
1955
1956                 if (cprm.limit < binfmt->min_coredump)
1957                         goto fail_unlock;
1958
1959                 cprm.file = filp_open(corename,
1960                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1961                                  0600);
1962                 if (IS_ERR(cprm.file))
1963                         goto fail_unlock;
1964
1965                 inode = cprm.file->f_path.dentry->d_inode;
1966                 if (inode->i_nlink > 1)
1967                         goto close_fail;
1968                 if (d_unhashed(cprm.file->f_path.dentry))
1969                         goto close_fail;
1970                 /*
1971                  * AK: actually i see no reason to not allow this for named
1972                  * pipes etc, but keep the previous behaviour for now.
1973                  */
1974                 if (!S_ISREG(inode->i_mode))
1975                         goto close_fail;
1976                 /*
1977                  * Dont allow local users get cute and trick others to coredump
1978                  * into their pre-created files.
1979                  */
1980                 if (inode->i_uid != current_fsuid())
1981                         goto close_fail;
1982                 if (!cprm.file->f_op || !cprm.file->f_op->write)
1983                         goto close_fail;
1984                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
1985                         goto close_fail;
1986         }
1987
1988         retval = binfmt->core_dump(&cprm);
1989         if (retval)
1990                 current->signal->group_exit_code |= 0x80;
1991
1992         if (ispipe && core_pipe_limit)
1993                 wait_for_dump_helpers(cprm.file);
1994 close_fail:
1995         if (cprm.file)
1996                 filp_close(cprm.file, NULL);
1997 fail_dropcount:
1998         if (ispipe)
1999                 atomic_dec(&core_dump_count);
2000 fail_unlock:
2001         coredump_finish(mm);
2002         revert_creds(old_cred);
2003 fail_creds:
2004         put_cred(cred);
2005 fail:
2006         return;
2007 }